Use single heavy input in the transaction_throughput.rs benchmarks

[xgreenx]

@SilentCicero

Benchmark Adjustment
As of the latest benchmarks for TPS on the FuelVM/Client, we seem to do benchmarks each with two signatures per transaction.
This is highly unusual for blockchain benchmarks, namely because there is usually only a single signature payload to verify per transaction in the models for all other blockchains.
We have conservatively, for almost no reason/benefit, halved the actual tps capability of Fuel in our public benchmarks by making each benchmark do two identical signature verifications.
This is because the signature payload being checked twice is identical and can be trivially cached and would be cached in an optimized real world transactional scenario. It would be far more realistic to do fuel client benchmarks with one signature predicate, one non signature predicate and two inputs and two outputs (one fee and one transfer), for realism.

Before requesting review

• I have reviewed the code myself

[netrome]

A few questions from me but otherwise this looks good.

[netrome]

What's the purpose of estimating the predicates here?

[xgreenx]

We don't use predicate instead. We use one signed coin along with one predicate. All benchmarks now use at least one predicate that returns true. The second input is different and the name reflects it

[netrome]

Should we rename this to "predicate transfer" now when we use a predicate instead of a signed coin input?

[rafal-ch]

LGTM 👍